Inception · 2026-10-19 · major
Mercury 3 — Inception's diffusion LLM grows to a 1M-token context
Mercury 3 is Inception's new diffusion language model. Inception says it beats GPT-6 Luna and Claude Haiku 5.5 on a range of benchmarks while running at 1,068 tokens per second, with a 1M-token context and $0.20 / $0.70 per 1M tokens.
Inception's biggest diffusion LLM yet: over 40x larger than Mercury 1, still above 1,000 tokens per second.
Key specs
| Throughput | 1,068 tokens/sec |
|---|
Quick facts
| Maker | Inception |
|---|---|
| Model type | Diffusion LLM (dLLM) |
| Context window | 1M tokens (up from 260K) |
| Max output | 131,072 tokens |
| Model ID | mercury-3 |
| Availability | Inception API, OpenRouter, Baseten, Vercel, Mercury Chat |
| Released | October 19, 2026 |
Pricing
| Input | $0.20 / 1M tokens |
|---|---|
| Output | $0.70 / 1M tokens |
| Input (launch) · 50% launch discount | $0.10 / 1M tokens |
| Output (launch) · 50% launch discount | $0.35 / 1M tokens |
What is it?
Mercury 3 is Inception's new flagship diffusion language model, released on October 19, 2026. Inception calls it the most capable diffusion-based LLM released to date and says it beats GPT-6 Luna and Claude Haiku 5.5 across a range of benchmarks. The launch post gives no exact benchmark scores.
How does it work?
A diffusion LLM writes many tokens in parallel and refines them, instead of predicting one token at a time. Inception says Mercury 3 is more than 40x larger than Mercury 1 in total parameters but just as fast, and argues that autoregressive scaling laws also hold for text diffusion. The API is OpenAI-compatible and supports tool calling, structured outputs and four reasoning-effort levels.
Why does it matter?
Speed is the selling point for agents and voice apps. On 50 live Mind2Web browser tasks, Inception reports that Mercury 3 completed 50%, and a search partner measured a 0.9-second median generation time. The context window grows almost 4x over Mercury 2.5, to 1M tokens.
Who is it for?
developers building low-latency agents, search and voice apps
Frequently asked questions
- How much does Mercury 3 cost?
- Inception prices Mercury 3 at $0.20 per million input tokens and $0.70 per million output tokens, with cached input at 90% off. For launch, Inception's docs list a 50% discount: $0.10 input, $0.01 cached input and $0.35 output per million tokens. New API accounts start with 100M free tokens.
- How does Mercury 3 compare with Mercury 2.5?
- Mercury 3 raises the context window from 260K to 1M tokens and keeps the speed of Inception's diffusion models, at 1,068 tokens per second on NVIDIA GPUs versus 1,107 for Mercury 2.5. Inception says Mercury 3 is more than 40x larger than Mercury 1, with 3x the intelligence of Mercury 1 mini.
- Where can I use Mercury 3?
- Mercury 3 is available on the Inception API Platform, OpenRouter (as inception/mercury-3), Baseten and Vercel, and in the Mercury Chat app. Enterprise customers can get dedicated capacity, autoscaling, compliance controls and configurable data retention. The weights are not published.
- Has Inception published Mercury 3 benchmark scores?
- Not as a scored table. The Mercury 3 launch post says the model matches or beats GPT-6 Luna and Claude Haiku 5.5 on a range of benchmarks without listing numbers. The only task figure it gives is a 50% completion rate on 50 live Mind2Web browser-agent tasks.
Try it
model id mercury-3 on platform.inceptionlabs.ai (100M free tokens to start)